Chujie Zheng
@ChujieZheng
Followers
6K
Following
2K
Media
66
Statuses
587
Researcher @Alibaba_Qwen | GSPO, Qwen3, QwQ, ProcessBench | Opinions are my own
Joined February 2018
Proud to introduce Group Sequence Policy Optimization (GSPO), our stable, efficient, and performant RL algorithm that powers the large-scale RL training of the latest Qwen3 models (Instruct, Coder, Thinking) 🚀. 📄
27
245
2K
RT @Alibaba_Qwen: 💡 You get 2,000 free Qwen Code runs every day!. Run this one simple command:.npx @qwen-code/qwen-code@latest.Hit Enter,….
0
536
0
Enjoy.
💡 You get 2,000 free Qwen Code runs every day!. Run this one simple command:.npx @qwen-code/qwen-code@latest.Hit Enter, and that’s it!.🚀 Now with Qwen OAuth support — super easy to use. Try it now and supercharge your vibe code! 💻⚡.Github：
0
4
36
RT @Alibaba_Qwen: 🚀 Qwen3-30B-A3B-2507 and Qwen3-235B-A22B-2507 now support ultra-long context—up to 1 million tokens!. 🔧 Powered by:. • Du….
0
520
0
Such short yet effective CoT. May stem from the extraordinary intuition. Where does the intuition come from? Guess the gigantic synthetic data.
1
0
35
LLM+RL feels like a new discipline, not just an extension of traditional RL. Truly fascinating to dive into its underlying principles and mechanisms.
13
69
1K
RT @Alibaba_Qwen: 🚀 Introducing Qwen3-4B-Instruct-2507 & Qwen3-4B-Thinking-2507 — smarter, sharper, and 256K-ready!. 🔹 Instruct: Boosted ge….
0
402
0
second, same feeling.
A hypothesis: gpt-oss is trained entirely on synthetic data, from pre-training to post-training. The approach enhances safety and helps smaller models achieve better performance.
1
0
21
cool.
0
1
19
A feast for the eyes, a shock to the soul. Try Qwen-Image now 🤗.
🚀 Meet Qwen-Image — a 20B MMDiT model for next-gen text-to-image generation. Especially strong at creating stunning graphic posters with native text. Now open-source. 🔍 Key Highlights:.🔹 SOTA text rendering — rivals GPT-4o in English, best-in-class for Chinese.🔹 In-pixel
1
0
21
RT @Alibaba_Qwen: 🚀 Meet Qwen-Image — a 20B MMDiT model for next-gen text-to-image generation. Especially strong at creating stunning graph….
0
670
0
RT @qingke_ai: 北京时间，8月7日晚8点，#青稞Talk 第68期，通义千问研究员，Qwen3、QwQ系列开源模型核心贡献者郑楚杰 @ChujieZheng ，将在 #青稞社区 直播分享《GSPO：迈向持续拓展的语言模型强化学习》。. #青稞社区 主页：https….
0
2
0
Uploaded on July 25, GSPO is already the #1 most popular paper on @huggingface for July 🍻
The GSPO paper by @Alibaba_Qwen is already the third most popular one on @huggingface for the month of July. I suspect this will have a massive impact on the field! Also, let's get back to celebrate research papers as massive contributions to the field?
2
15
162
Congrats to our amazing team @Alibaba_Qwen 🍻.
🚨BREAKING: @Alibaba_Qwen’s latest Qwen3 now claims the #1 open model in the Arena!. With 3k+ community votes, it ranks #3 Overall and ties for #1 in Coding, Hard Prompts, and Math - overtaking DeepSeek and Kimi-K2 as the top open model. Huge congrats to the Qwen team on this
0
1
21
RT @lmarena_ai: 🚨BREAKING: @Alibaba_Qwen’s latest Qwen3 now claims the #1 open model in the Arena!. With 3k+ community votes, it ranks #3 O….
0
74
0
RT @cherry_cc12: Excited to announce that ms-swift has integrated GSPO! 🚀 Now available via --importance_sampling_level sequence. Check ou….
github.com
PR type Bug Fix New Feature Document Updates More Models or Datasets Support PR information Write the detail information belongs to this PR. Experiment results Paste your experiment result he...
0
2
0
Coder flash with amazing power!.
🦥 Qwen3-Coder-Flash: Qwen3-Coder-30B-A3B-Instruct.💚 Just lightning-fast, accurate code generation. ✅ Native 256K context (supports up to 1M tokens with YaRN).✅ Optimized for platforms like Qwen Code, Cline, Roo Code, Kilo Code, etc. ✅ Seamless function calling & agent
2
0
14
RT @Alibaba_Qwen: 🦥 Qwen3-Coder-Flash: Qwen3-Coder-30B-A3B-Instruct.💚 Just lightning-fast, accurate code generation. ✅ Native 256K context….
0
441
0
RT @huybery: Since the release of Qwen3-Coder a week ago, it has got a lot of love from the community! We have received many valuable sugge….
0
54
0
@verl_project Also has been integrated into OpenRLHF (, ROLL ( and slime (. Now these mainstream RL frameworks (all from China!) all have supported GSPO! 🥳.
github.com
cc @ocss884
0
2
8
GSPO has been integrated into @verl_project ( and TRL (. Thanks for the prompt support from the community 🚀.
github.com
Breaking and major changes 🎞️ GSPO GSPO is a GRPO variant that computes importance sampling weights at the sequence level instead of per-token. 📜 Paper: https://huggingface.co/papers/2507.18071...
Proud to introduce Group Sequence Policy Optimization (GSPO), our stable, efficient, and performant RL algorithm that powers the large-scale RL training of the latest Qwen3 models (Instruct, Coder, Thinking) 🚀. 📄
3
24
204